I judge every short by its first three seconds in isolation, because that is the only part a stranger actually watches before deciding whether to stay.
I edit short-form video the same way every time: I cut the first three seconds loose from the rest of the piece and watch them on their own, with no idea what comes next, and ask whether a stranger scrolling past would stop. If the answer is no, I do not try to rescue it later in the edit. I reshoot the open, because nothing after that point gets a chance to matter.
This sounds obvious once you say it, and it is still the mistake I catch most often in my own footage. The instinct when reviewing a full cut is to judge it as a whole — does the argument hold together, does the ending land — and that is exactly the judgment a viewer never makes. A viewer decides whether to keep watching before any argument has had time to form. Whatever earns the stay has to work with no context at all, standing completely alone. I have tested this against my own reaction as a viewer, scrolling my own feed with the sound off, and the pattern holds without exception — the pieces that survive the first beat are never the best-argued ones, they are the ones with something true and slightly unexpected sitting right at the front.
The opening line I reach for on a vyke.ai short is not a hook in the marketing sense, a tease or a question dangling a promise. It is a claim that is already interesting on its own, stated flat, with the proof visibly still owed. "I ran the same prompt on four models and only one of them told me it couldn’t do it" works because it is a complete, surprising sentence by itself — it does not need the rest of the video to mean something, it only needs the rest of the video to back it up.
Under that sits a pacing rule I think about as value-per-line rather than value-per-video. Every sentence in the script has to earn its own place, not just contribute to an overall theme building towards a payoff at the end. I go through a script line by line and ask, of each one, whether it could be cut with nothing lost — if the answer is yes, it goes, because a viewer who has not committed to the whole piece is judging each line on its own, not banking on a slow build. This is a harder edit than it sounds, because a line doing real work in the middle of a longer piece often reads as unnecessary in isolation, and cutting it changes the argument enough that two or three surrounding lines have to be rewritten to hold without it.
The failure mode I see most often in other people’s shorts, and used to make constantly in my own, is front-loading the setup instead of the claim. Explaining who you are, what you are about to show, why it matters — all of that is true information and none of it earns three seconds. The claim has to come before the context, every time, even when it feels rude to skip the introduction.
None of this is really about video. It is about writing for an audience that has not yet decided to trust you, which is a different discipline from writing for one that has. Most long-form writing advice assumes the reader is already staying; short-form retention assumes they are one beat from leaving, and every piece of craft downstream of that assumption looks different once you take it seriously. The two disciplines are not opposed, but treating short-form retention as a lesser version of long-form writing, rather than a genuinely different skill with its own rules, is the fastest way to keep producing openings that read well on the page and get skipped on the feed.
I keep a running note of openings that failed this test, and rereading it is more useful than any list of ones that worked, because the failures cluster around the same handful of habits — a scene-setting shot before the claim, a name or a logo before the claim, a question before the claim — and every one of those habits is a small act of politeness that costs the only three seconds I actually get.